We were once deeply troubled by getting correct data out of documents. You don’t have to be. A wrong load rating, a shifted column, an invented part number — critical data errors have serious consequences long after the spreadsheet looks fine. And throwing an AI model at it doesn’t fix this: even the best models hallucinate, inventing plausible values and stamping them “high confidence”. That’s why nothing here is trusted until independent readings agree. Consensis is the tool we made for our own catalogues — now we’re sharing it.

— why we’re sharing Consensis

Measured, not claimed.

We score ourselves against hand-built, page-level ground truth — built before running any tool, never sampled. The same open scoring script is applied to us and to the competition.

Metric	Consensis	Leading competitor*
Cell values correct	96%	92%
Whole rows correct	94%	75%
Table structure correct	76%	58%
Grid consistency	100%	99%
No invented values	99%	97%

* A leading VLM-based extraction API, run in its highest-accuracy agentic mode, outputs frozen before our re-runs.

Four readers. No single point of trust.

Accuracy doesn’t come from a bigger model — it comes from independent readings that must agree before a value is trusted.

Upload

Send a PDF through the API or the browser. Catalogues, datasheets, price lists — text layer or scan.

Read independently

Multiple OCR engines, a geometry parser and a layout model each extract every page, blind to each other.

Cross-check per cell

Readings are compared cell by cell. Agreement becomes verified; disagreement ships with every version and the original crop.

Deliver & review

JSON or Excel, with per-cell source and confidence. A side-by-side review page for any page you want to check yourself.

Fix it online — no API, no code.

Every job ships with an interactive review page. Open it in your browser, and every flagged cell is one click away from settled — side-by-side with the original page. Work through the short flagged list yourself — most documents come out with only a handful.

See the evidence

Click a flagged cell and the original page zooms to the exact spot, with every reader's version listed.

Confirm or correct in place

The page is right? One click. It isn't? Type the fix — subscripts and engineering symbols included — and move on. A “next to review” button walks you through what's left.

Export the corrected data

Excel or CSV for spreadsheets — or one ZIP with the corrected JSON and every cropped image, enough to rebuild each page. Your edits stay in your browser; we never see them.

Check this value against the highlighted cell on the original page. Fix it here if it’s wrong, then press Confirm.

On the original page: 18.8 ✓ Confirm

X₂ subscriptX² superscript °±×≤Ø

Your documents stay yours.

Catalogues carry pricing and unreleased products. We treat retention as a liability, not an asset.

Deleted in 24 hours

Results, crops — and the OCR cache — are wiped 24 hours after completion. Nothing is kept to train on or resell.

Delete it now

DELETE /v1/parse/{id} removes a job immediately — you don’t have to wait for the window.

No hidden copies

Once deleted, a resent document is processed — and billed — as a brand-new job. Keeping a shadow copy would be cheaper for us; we don’t.

Pricing

$0.06 / delivered page

Batch mode — $0.036 / page: same extraction and the same guarantees, results in 10–60 minutes instead of seconds per page. Pick it per job with mode=batch
Free trial — 10 pages a day on the website and another 10 on the API, no account needed
Prepaid credits — minimum top-up US$3.50 (58 pages, or ≈97 in batch)
No subscription, no platform fee
Failed pages are never charged
Every job itemised on your account page

Create an account

Consensis is in beta — sign in with Google or email and you’re in. Questions, suggestions, volume and per-project quotes: [email protected] — no account needed to write to us.

The most accurate way to turn catalogue PDFs into data.

Consensis in 40 seconds